Gant Laborde
Gant Laborde
Can you provide the image you used? And yes, false positives will happen.
@TechnikEmpire - would know what's up. https://github.com/TechnikEmpire
Good call. We're not actively adding features at this time, but this would def be a good feature.
Agreed, this sounds like a separate project. If this is something you can either pay for ([email protected]) or get a large following to demand this as a product idea and...
You'll want to make a porn/not_porn group of roughly the same size. It's 2 categories. Preferably have the not_porn folder have a lot of images of people interacting, but not...
I think you need to get 10,000 non-porn photos. Then use the training code you have, but change the final layer to softmax sigmoid with 2 neurons. Porn and not-porn.
ACTUALLY, you could use "K-Nearest Neighbors Classifier" to turn the model into classifying 2 different items. If you use the inference data. I'm not an expert on how to do...
Yup, the transfer learning from this code would serve as a great tutorial for a different image recognition initiative. I only wish my code was cleaner :D
https://js.tensorflow.org/api/latest/#io.listModels
This should help: https://github.com/GantMan/nsfw_model